作者：链上观

我之前在多篇文章中说过AI Agent会是诸多Crypto行业老叙事的“救赎”。 In the last wave of narrative evolution around AI independence, TEE was once praised as the top spot. However, there is also a more "unpopular" technical concept than TEE or even ZKP. FHE - fully homomorphic encryption will also be "rebirth" due to the driving force of the AI ​​track. Below, let’s sort out the logic for you through cases:

FHE is a cryptography technology that allows direct calculations on encrypted data. It is regarded as "Holy Grail". Compared with popular technology narratives such as ZKP and TEE, it is in a relatively unpopular position, and the core is mainly trapped by overhead and application scenarios.

Mind Network focuses on FHE infrastructure and launched FHE Chain, which focuses on AI Agent. Although it has raised more than 10 million US dollars and has experienced several years of technological hard work, due to FHE itself, market attention is still underestimated.

However, Mind Network has recently launched a lot of good news around AI application scenarios. For example, the FHE Rust SDK it developed is integrated by the open source big model DeepSeek, becoming a key link in the AI ​​training scenario, providing a security foundation for the implementation of trusted AI.为何FHE可以在AI 隐私计算中有所表现，能否借助AI Agent的叙事实现一次弯道超车 or 救赎吗？

Simple: FHE fully homomorphic encryption is a cryptographic technology that can directly act on the current public chain architecture, allowing for arbitrary calculations such as addition and multiplication on encrypted data without the need to decrypt the data first.

In other words, the application of FHE technology can enable full encryption of data from input to output, and even nodes that maintain public chain consensus for verification cannot access plain text information. In this way, FHE can provide technical underlying guarantees for the training of some AI LLMs in vertical segmentation scenarios such as medical care and finance.

让FHE可以成为传统AI 大模型训练丰富扩展垂直场景以及结合区块链分布式架构的一种“优选”解决方案。 Whether it is cross-institutional collaboration of medical data,Or the privacy reasoning of financial transaction scenarios, FHE can become a complementary choice with its uniqueness.

This is actually not abstract. You can understand it with a simple example: For example, as a C-end application, the background of the AI ​​Agent is usually connected to AI big models provided by different suppliers such as DeepSeek, Claude, OpenAI, etc., but how to ensure that in some highly sensitive financial application scenarios, the execution process of the AI ​​Agent will not be affected by the background of the big model that suddenly tampers with rules? This requires the input Prompt to be encrypted. When the LLMs service provider directly calculates the ciphertext, there will be no forced interference and changes that affect fairness.

So what's going on with the other concept of "trusted AI"? Trusted AI is a FHE decentralized AI vision that Mind Network is trying to build, including allowing multiple parties to achieve efficient model training and inference through distributed computing power GPUs, without having to rely on a central server, and providing FHE-based consensus verification for AI Agents. This design eliminates the limitations of originally centralized AI and provides dual guarantees of privacy + autonomy for the operation of web3 AI Agent under a distributed architecture.

This is more in line with the narrative direction of Mind Network's own distributed public chain architecture. For example, in special on-chain transactions, FHE can protect the privacy reasoning and execution of Oracle data of all parties, allowing AI Agents to make independent decisions on transactions without exposing positions or strategies, etc.

So, why does FHE have similar industry penetration paths like TEE, which will bring direct opportunities due to the outbreak of AI application scenarios?

Previously, TEE was able to seize the opportunity of AI Agent because the TEE hardware environment could enable data to be hosted in a private state, and then allow AI Agent to independently host private keys, allowing AI Agent to reach a new narrative of independent management of assets. However, TEE actually has a flaw in keeping private keys: trust depends on third-party hardware providers (such as Intel). To make TEE work, a distributed chain architecture is needed to add an additional open and transparent "consensus" constraint to the TEEs environment. In contrast, PHE can exist entirely based on a decentralized chain architecture without relying on third parties.

FHE and TEE have similar ecological nichesAlthough TEE is not widely used in the web3 ecosystem, it has long been a very mature technology in the web2 field. In contrast, FHE will gradually find value in both web2 and web3 under the explosion of this round of AI trends.

Above.

To sum up, it can be seen that FHE's encryption Holy Grail-level encryption technology, under the premise that AI becomes the future, will inevitably become one of the cornerstones of security and be further widely adopted.

Of course, despite this, it is necessary to avoid the overhead cost issue of FHE when implementing algorithms. If it can be applied in the web2 AI scenario and then linked to the web3 AI scenario, it will probably unexpectedly release the "scale effect" and dilute the overall cost, so that it can be more popular.